A quick note: I missed that the Cards-Rams game in week one was a push, so I updated that in my week one round-up post. Not that it matters much here because that was based on a different model, but it makes both Bill Simmons and I look a bit better. Before we get to the predictions, a few reminders: I’ll post predictions after kick-off. I’ll likely be going to trivia before the Sunday night game starts, so I’ll include that in the afternoon predictions and you can all have a freebie. But, you’ll need to check this post again a couple times to get the updates: just after noon (Central time), again around 3:20, then again Monday night. I’m going to post predictions from two models, each of which predict win probability, point difference, and point total (they’re described a bit more in the NFL model link under the banner). Given the past two years, I’m expecting the big model to do better against the line and the other one to do better with the win and over/under predictions. I’m also going to list Bill Simmons’ picks against the spread as a comparison. I’ll use both his lines, because I don’t know where they come from, and the lines, moneyline, and over/under from Bodog, which I’ll get Sunday morning.
Finally, some of the numbers this week will look funny. That’s because there’s only one game to base them on, so there are some relatively extreme values being fed in. But, that one game is informative. The past two years, the big model (I’ll call it Mario, because it’s the money maker) has gone 19 of 31 with one push (19-11-1), and the middle model (Luigi) has gone 16 of 31 with a push (16-14-1) in week 2 (But, much like the stock market, past results do not guarantee future returns). Also, I haven’t tracked the moneyline or over/under historically, so I can’t say much as to how to pick those exactly; at the end of the year I might have enough data to evaluate the accuracy of the models. But, without further ado, here are the picks:
There are three games that don’t have a moneyline posted; they’re listed as 0. Remember that the ‘pick’ here isn’t the winner (you can get that straight from the probabilities), it’s if the home team is a good pick given their moneyline. Unlike the over/under and the spread pay odds, which sum to -220, the moneyline for home and away don’t add to anything in particular, so in the future I’ll need to record both of those.
The models are in good agreement again except for three games. You might notice that the spread picks are inconsistent with the winner picks in some cases. For example, Mario thinks that the Packers only have a 36.6% chance of winning, but should win by 4.74 points. That’s because each of the three predictions (winner, spread, over/under) are actually three different models; when I say I have two models, I mean two sets of predictors. But these two sets are predicting three different dependent variables, so the weights will move around and they may not be consistent.
Finally, the Bill Simmons picks. You can see we have some disagreements, where neither model agrees with Bill. We’ll see how it goes.